Background in Speech Recognition

Abstract

The speech recognition process is usually described within the context of a communication problem. As it is shown in Figure 2.1, the ideas that the user wants to express are first mapped into a sequence of words W with a word generator. The sequence of words is encoded into a speech signal through the speech generator which consists of the vocal tract and the respiratory system. After that, the speech signal passes through a communication channel (air) and then it is converted to electrical signals via a microphone. This channel can be modeled with the transfer function h(t) and possible ambient sounds as additive noise n(t). Finally, the speech decoder receives the speech signal. The feature extractor module is in charge of taking out characteristics features from the signal and delivering them to the statistical decoder whose task is to estimate the sequence of words that the user has transmitted.